"What is shadow AI?" Owners usually ask it the week they find out. A customer list turns up in a free writing tool, a quote goes out that nobody in the office remembers writing, or a departing employee's personal account turns out to hold half the sales team's templates. The term sounds like a security problem, and the people selling detection software would like you to think of it that way. For the owner of a company with fifty to five hundred people it is something plainer: the AI already in use in your company that nobody approved and nobody wrote down.
I was COO and then President of a public company that closed more than forty acquisitions, and I ran the integration on most of them. In every company we bought, the tools people actually used were never the tools on the approved list, and the people using them were usually the ones worth keeping. I am a lawyer by training. Now I write fixed-fee AI plans for CEOs of private-equity-backed and owner-led mid-market companies through 013 Labs, and the first thing I do in any of them is find the shadow AI, because it tells me more about the company than any org chart. Not legal advice; a working owner's view.
The short answer
Shadow AI is any AI tool your people use for company work that the company did not choose, does not pay for, and has no written rule about. A free account opened with a personal email. A feature a vendor switched on inside software you already own. A browser add-on that reads every page. A chat tool on a phone that has your pricing in it because someone wanted a faster answer on a Friday afternoon. It is in almost every company of this size today, and the owner is usually the last to know.
It is not a crime, and treating it as one is the most common mistake. People reach for these tools because something in their job is slow, dull or late, and because nobody told them what they were allowed to do. The ones who went first are showing you where the work hurts and who in the building has the initiative to fix it. That is exactly the person the one-page guide, Spot your AI champion, is written to find. The problem is not that they used a tool. The problem is that nobody knows what went into it, under what terms, and what came out signed in the company's name.
What shadow AI looks like in a company this size
It rarely looks like a rogue department. It looks like ordinary people being helpful. The estimator who pastes a drawing and a bill of materials into a free tool to get a first-pass quote. The customer service lead who keeps a chat window open to soften difficult emails, and pastes the customer's whole thread in to do it. The sales rep who built a prospecting list in a personal account that will leave with them. The operations manager who turned on the summary feature in the meeting software without asking where the recordings go.
The common thread is that company information moved to a place the company does not control, and that the person doing it thought they were doing their job well. Usually they were. Who owns AI in a company explains why, in companies this size, nobody is in charge of these decisions by default, and shadow AI is simply what happens in that gap.
How to find it in a week, without a witch hunt
You do not need software to find shadow AI. You need to ask, in a way that makes it safe to answer. Tell the company plainly, in person or in a short note, that you expect people are already using these tools, that you are glad they are, and that for the next week there is no penalty for saying so. Then ask each manager for one short list from their team: which tools, for what task, with what kind of information. Most of the list arrives within days, because people have been waiting to be asked.
Do not start by searching laptops or pulling browser histories: it tells your best people that initiative is punished and pushes the real usage onto phones, and the list you get by asking comes with the reasons attached, which are the valuable part.
Keep the list short and plain: tool, task, type of information, who uses it, paid or free, company account or personal. One page for the whole company is normal. Then sort it into three piles: tools that handled nothing sensitive and can carry on, tools that touched customer, employee or financial information and need a decision, and tools holding company work in someone's personal account that must be moved before anything else.
The three lines that end it as a problem
Shadow AI stops being a problem the day three lines exist in writing and everyone has seen them. They fit on half a page.
What never goes into an outside tool. Name the categories in words your people use: customer lists and contact details, prices and quotes before they are sent, payroll and anything about a named employee, contracts, drawings and recipes, bank and payment details, and anything a customer gave you in confidence. One sentence per category is enough.
Which tools are allowed, on a company account. A short list of the tools the company pays for or has chosen, each on a company login the company controls, with written terms that say who keeps the data and what it is used for. Anything not on the list is not banned; it is "ask first". That one word keeps the door open for the next good idea.
Who to ask. One named person, usually not the owner, who can say yes to a new tool within a day and knows what to check before saying it. Often this is the person who was already using the tools quietly. Giving them the question in public turns the problem into the answer.
The policy that grows from these three lines is covered in Do I need an AI policy for my small business, and the wider question of what the company decides before any of this is in Do I need an AI strategy. Not legal advice; a policy that touches employment or regulated data should be checked by someone qualified to check it.
The data rule
The first of the three lines deserves its own paragraph, because it protects the company whether or not the other two are ever written. Before anyone puts customer records, payroll, contracts or pricing into any tool, the company needs one short rule: which tools are allowed, under written terms that say who keeps the data and what it is used for, never through a free personal account, and with a named person who may approve a new one. Free tools are free because the terms suit the provider, not you. What happens to your data sets out what to ask any provider before the first file moves. Write the rule before the next quote goes out, not after the first scare.
Five checks on whether it is handled
First, ask three people at random what they may not put into an outside tool. If the three answers differ, the first line has not landed.
Second, count the personal accounts. Any company work still living in a personal login is a resignation letter away from leaving with its owner.
Third, find the name. Someone in the building can say yes to a new tool within a day. If the answer is "the owner", the owner is the bottleneck and the shadow will return.
Fourth, look at the last quote or contract that a tool helped draft. A person checked it before it went out, and can say so. If not, the tool is signing in the company's name.
Fifth, listen for the word banned. If people describe the rule as a ban, they will work around it. If they describe it as "ask first", they will ask.
Check yourself
Seven yes or no questions. Each no is a decision still to make.
- Could you name, today, five AI tools your people are already using for company work?
- Has anyone asked them, in a way that made it safe to say yes?
- Is there a one-sentence list of what never goes into an outside tool, and has everyone seen it?
- Is every tool that touches customer or employee information on a company account with written terms?
- Is there one named person, not you, who can approve a new tool within a day?
- Does a person read anything a tool drafts before it goes to a customer with the company's name on it?
- Would company work survive the departure of your most enthusiastic tool user?
Mostly no? That is normal, and it is a week of asking, not a year of policing. The AI readiness scorecard asks ten questions like these across the whole company.
Starting safely
Start by saying, in public, that you know people are using these tools and that you are glad. That one sentence does more than any policy. Then ask the managers for the lists, write the three lines in words your warehouse would understand, and put them where people will see them, not in a handbook nobody opens. Move personal-account work first, because it is the only piece that gets worse with time. Choose the named person from among the early users, not from among the cautious.
What not to do is just as short: do not ban everything, do not buy a monitoring tool before you have asked, do not punish the people who went first, and do not write twelve pages to solve a half-page problem. How to train employees to use AI covers what to do once the three lines exist, and Is ChatGPT safe for business use answers the question most of them will ask first.
Picture it going wrong
A made-up example, to show the shape. A family-owned industrial distributor with about one hundred and twenty people and a long customer list. The owner reads about shadow AI in a trade newsletter, is alarmed, and has the IT contractor block the main chat tools at the office firewall the same afternoon, with a company-wide email saying AI use is now prohibited pending a policy. Within a week the estimating team is doing the same work on their phones, the customer service lead has moved her templates to a personal account at home so she can keep answering emails quickly, and the one person who had built a genuinely useful parts-lookup helper stops mentioning it. Three months later a departing rep takes a personal account with four years of customer correspondence in it, and the owner learns about the parts-lookup helper from a customer who liked it. The company lost sight of its own best work and its own information in the same month, because the first reaction was a ban instead of a question.
Mistakes that repeat
- Treating shadow AI as a security incident instead of a sign of where the work hurts.
- Banning first and asking never, so the usage moves to phones and personal accounts.
- Buying detection software before anyone has simply asked the team.
- Letting company work accumulate in personal accounts because moving it feels like an accusation.
- Writing the three lines and filing them in the handbook instead of saying them out loud.
- Making the owner the only person who can say yes, so every request waits and the shadow returns. Not accounting or tax advice where the tools touch the books.
Frequently asked questions
What is shadow AI, in plain words?
Any AI tool your people use for company work that the company did not choose, does not control and has no written rule about: free chat tools on personal accounts, add-ons in a browser, and features vendors switch on inside software you already pay for.
Is shadow AI dangerous?
It can be, in three specific ways: company or customer information leaves under terms nobody read; a quote, email or contract drafted by a tool goes out unread in the company's name; and company work ends up in a personal account that leaves with the person. All three are fixed by the three lines on this page, not by a ban.
How do I find out which AI tools my employees are using?
Ask. Say in public that you expect it, that you are glad, and that there is no penalty for a week. Have each manager send one short list: tool, task, type of information, company account or personal. You will have most of the picture in days without touching a single laptop.
Should I ban ChatGPT and similar tools at work?
Almost never. A ban moves the usage to phones and personal accounts, where you cannot see it, and tells your most capable people that initiative is punished. Allow a short list on company accounts, say what may never go in, and make everything else "ask first".
Who should own this in a company with no IT department?
One named person who already uses the tools well and is trusted by their colleagues, with a written scope and the authority to say yes within a day. The owner keeps the data rule and the decision about which tools the company pays for.
What should an AI policy say about shadow AI?
Three things: what never goes into an outside tool, which tools are allowed and on what accounts, and who to ask. A half page that everyone has seen beats a long policy nobody reads. Anything touching employment or regulated data should be checked by someone qualified; this is not legal advice.
How do I know the problem is handled?
When the list of tools in use is written and changing, personal-account work has been moved to company accounts, the sensitive tools are either on proper terms or retired, and the named person has said yes to something new. Thirty days is enough to see all four.
Not sure where you stand? Take the 3-minute AI readiness scorecard.